Transistors: On or Off

It's Monday, May eighteenth. The phone in your pocket contains more computing power than every computer that existed on Earth in 1960 combined. That computing power runs on a chip the size of your thumbnail, packed with billions of transistors, each one smaller than a virus. The transistor, a tiny switch that's either on or off, one or zero, is the most manufactured object in human history. Roughly thirteen sextillion of them have been produced. That's more transistors than all the grains of sand on all the beaches on Earth. We're covering this today because semiconductors are the foundation of everything digital, from your phone to your car to the AI models that are reshaping entire industries, and most people couldn't explain what one actually does. The story of how we went from a single transistor in a New Jersey lab in 1947 to a chip with a hundred billion of them involves the hardest manufacturing process humans have ever attempted, one company in the Netherlands that makes the only machine capable of doing it, and one island in the Pacific that the entire global economy depends on. In December 1947, three physicists at Bell Labs in New Jersey, John Bardeen, Walter Brattain, and William Shockley, built the first working transistor. It was about the size of a pencil eraser, held together with a paperclip and a strip of gold foil. It replaced the vacuum tube, which did the same job but was the size of a light bulb, generated enormous heat, and burned out constantly. The ENIAC computer, built a year earlier, used 17,468 vacuum tubes, filled an entire room, and required a full-time staff just to replace the tubes that failed. The transistor did the same work at a fraction of the size, with almost no heat and almost no failure rate. Bardeen, Brattain, and Shockley won the Nobel Prize in 1956. A transistor is, at its simplest, a switch. It can be on or off. One or zero. Every computation your phone performs, every photo it displays, every message it sends, is ultimately a sequence of ones and zeros processed through billions of these tiny switches flipping on and off billions of times per second. The magic isn't in any individual switch. It's in the speed and the scale. In 1958, Jack Kilby at Texas Instruments and Robert Noyce at Fairchild Semiconductor independently figured out how to put multiple transistors on a single piece of silicon. The integrated circuit, or chip, was born. Instead of wiring individual transistors together by hand, you could print them onto silicon in a single manufacturing step. In 1965, Gordon Moore, who co-founded Intel with Noyce, observed that the number of transistors on a chip was doubling roughly every two years. This observation, which became known as Moore's Law, held for six decades. The first Intel processor in 1971 had 2,300 transistors. Apple's M4 chip has over 28 billion. The doubling happened so consistently, for so long, that it became the metronome of the entire technology industry. Every product roadmap, every R&D investment, every business plan assumed the chips would keep getting smaller, faster, and cheaper. And they did. Making a modern chip is the most complex manufacturing process in human history. The basic method is called photolithography: you project a pattern of light onto a silicon wafer coated with a light-sensitive chemical, and the light carves the pattern of transistors into the surface. The problem is that modern transistors are so small, about three nanometers wide on the most advanced chips, that the wavelength of visible light is too large to carve them. It's like trying to paint a portrait with a fire hose. The solution came from a Dutch company called ASML, and it took decades to develop. ASML builds machines that use extreme ultraviolet light, or EUV, a wavelength so short it's almost an X-ray, to carve patterns onto silicon at a scale that would have been considered physically impossible twenty years ago. Each EUV machine has over 100,000 parts. It weighs roughly 180 tons and ships in 40 freight containers. The latest versions cost over $380 million each. ASML is the only company on Earth that makes them. Every advanced chip in every phone, laptop, data center, and AI system in the world was manufactured using an ASML machine. If ASML's factory in Veldhoven, Netherlands, stopped producing tomorrow, the global technology industry would grind to a halt within months. The companies that use ASML's machines to actually manufacture chips are called foundries. The largest is Taiwan Semiconductor Manufacturing Company, TSMC, which produces roughly ninety percent of the world's most advanced chips. TSMC doesn't design chips. Companies like Apple, Nvidia, AMD, and Qualcomm design them. TSMC manufactures them. This split, where the designers and the manufacturers are completely separate companies, was pioneered by TSMC's founder Morris Chang in 1987. It's called the fabless model, and it reshaped the entire industry. Before TSMC, if you wanted to design a chip, you also had to build and operate a multibillion-dollar factory. After TSMC, you just sent your design to Taiwan and they built it for you. That model unlocked an explosion of chip innovation because it lowered the barrier to entry for designers. It also concentrated manufacturing in one place to a degree that makes military strategists nervous. If you've heard the name Nvidia in the past few years, it's probably in connection with artificial intelligence. Nvidia is now one of the most valuable companies on Earth, worth over three trillion dollars. But Nvidia wasn't built for AI. It was built for video games. In 1999, Nvidia released the GeForce 256, which it called the world's first GPU, or graphics processing unit. The chip was designed to handle the complex visual calculations required for 3D gaming: rendering lighting, shadows, explosions, textures. To do this, GPUs were designed differently from CPUs. A CPU has a few powerful cores that handle tasks one at a time, very fast. A GPU has thousands of simpler cores that handle many tasks simultaneously. A CPU is a brilliant single worker. A GPU is a massive team of average workers doing the same job in parallel. In 2012, a researcher at the University of Toronto named Alex Krizhevsky used two Nvidia gaming GPUs to build a neural network that crushed the competition at ImageNet, a major image recognition contest. The AI community noticed. It turned out that training a neural network, which involves adjusting millions or billions of numerical parameters through repeated calculations, is almost exactly the same kind of math that rendering explosions in a video game requires: millions of simple operations happening simultaneously. The GPU wasn't designed for AI. It was accidentally perfect for it. Nvidia recognized what was happening before anyone else. In 2006, it had already released CUDA, a software toolkit that let developers use GPUs for general-purpose computing beyond graphics. When the AI boom arrived, Nvidia had both the hardware and the software ecosystem ready. Its data center revenue, which barely existed a decade ago, now accounts for over ninety percent of its total revenue. The gaming business that built the company has become a sideshow. Gamers who once made Nvidia profitable now complain that the company has abandoned them for AI customers willing to pay far more. "Dance with the one who brought you," said one gaming podcast host. Nvidia, clearly, is dancing with someone else. The CHIPS and Science Act, signed in 2022, allocated $280 billion to bring semiconductor manufacturing back to the United States. TSMC is building factories in Arizona. Intel is expanding in Ohio and Oregon. Samsung is building in Texas. The factories need tens of thousands of workers, and many of the roles, technician positions in fabrication and equipment maintenance, require a two-year degree or technical training rather than a four-year engineering degree. The chip industry is one of the few sectors where manufacturing jobs are being created in the US rather than moved overseas, and the entry point is more accessible than most people assume. So if this comes up in conversation, here's how to think about it. A transistor is a tiny switch, on or off, one or zero. The first one was built in 1947. A modern chip has billions of them. Manufacturing these chips requires machines that cost $380 million each, made by one company in the Netherlands, used primarily by one company in Taiwan that produces ninety percent of the world's most advanced chips. Nvidia, originally a gaming company, accidentally became the engine of the AI revolution because the math behind rendering video game graphics turned out to be the same math behind training neural networks. The entire digital world, your phone, your car, the AI you talk to, runs on these chips, and the supply chain that produces them passes through a handful of irreplaceable chokepoints. The most important object in the modern world is something most people have never seen and couldn't explain. Now you can. Stay informed, stay curious, and we'll see you tomorrow.